Generative model based frame generation of volcanic flow video

Automatic generation of computer graphics utilizing generative models has been the state of the art recently. The motivation of generating images on a natural phenomenon with a generative model is to simulate it more easily than with conventional methods, which usually propose some mathematical equations for simulation. In this paper, we focus on generating volcanic flow images by utilizing Deep Learning methods, such as Deep Convolutional Generative Adversarial Networks (DCGAN) and Variational Autoencoders (VAE). In order to simulate lava flow, we adopt videos that are uploaded YouTube as input dataset. The experimental results demonstrate that using DCGAN to our problem is inferior to utilizing VAE in terms of time complexity and quality of output images. These results suggest that when input dataset has categories that contain sequential images, it is more effective to utilize VAE rather than to use DCGAN.